Jumpy is a common backend for NumPy and optionally JAX:
- If Jax is installed and jax inputs are provided then the
jax.numpyfunction is run - If Jax is installed and the function is jitted then the
jax.numpyfunction is run - Otherwise the jumpy function returns the NumPy outputs
There are several functions (e.g. vmap, scan) that available with jax installed.
Jumpy lets you write framework-agnostic code that is easy to debug by running as raw Numpy, but is just as performant as JAX when jitted.
We maintain this repository primarily so to enable writing Gymnasium and PettingZoo wrappers that can be applied to both standard NumPy or hardware accelerated Jax based environments, however this package can be used for many more things.
To install Jumpy from pypi: pip install jax-jumpy[jax] will include jax while pip install jax-jumpy will not include jax.
Alternatively, to install Jumpy from source, clone this repo, cd to it, and then: pip install .
Jumpy does not have a complete implementation of all numpy or jax.numpy functions.
If you are missing functions then please create an issue or pull request, we will be happy to add them.
In the future, we are interested in adding optional support for PyTorch and looking for pull request to complete this.